liquid nitrogen cooling a possibility?

Velocet math at velocet.ca
Tue Jun 12 07:41:52 PDT 2001


On Tue, Jun 12, 2001 at 11:54:59AM +0200, Frank Joerdens's all...
> I am trying to find information on whether anyone has tried to increase
> an off-the-shelf computer's performance by cooling either the CPU or the
> entire motherboard (probably more practical in terms of avoiding condensation
> issues) in liquid nitrogen. Are there components on a motherboard that

*IN*?

> will stop working (e.g. capacitors) or is this feasible? I have no idea

yes, at a certain temperature, semiconductors become conductors. The
lower the temp the more likely. not sure at what temp this is though.

You really dont need to run your components *AT* -180C+ odd whatever LN is at.
You just need to run them below 50C.  Actually, from what I understand, faster
speeds are more possible due to cleaner clock signals at lower temp (is this
true? thermal noise cant account for this - I was just reading Feynman's
famous speech in Japan about dropping the energy requirements of transistors
to get closer to electron thermal noise levels which is a dozen odd magnitudes
of lower energy than we still employ many years after his speech) - or perhaps
the clock just generates cleaner signals at those temps.

Starting the system would be a problem at those low temperatures. You'd need
to turn up the cooling system as heat started being generated, to ensure you
dont dip below some magical cutoff temp. And there are some other analog
components on mainbaords which may have acceptable parameters at -20C to 60C
operating temps, but will behave very oddly outside of those. A copper cooling
jacket for a water system supplanted with LN as the coolant for cooling the
CPU may work, but I suggest keeping a close eye on the CPU temp monitor,
however accurate it is. Not to mention if you heat the LN too much for such a
system, it becomes a gas and may exert signifcant pressure on whatever tubing
you are using.... A system with a variable speed pump may work well. Of course
this pump will have to be able to operate at incredibly cold temperatures.

Most mainboards have some sort of cutoff circuit from what I understand
if there is a large current flow through the cpu - I wouldnt rely
on this nonetheless as the temperature drops below -50C and lower,
however. You may well 'fry' it even tho its quite cold.

> whether this would be practical or desirable; at the moment my interest
> is academic. I thought if anyone knows about efforts in this area, it
> would be on this list . . . also, if it works it's probably hideously
> expensive for a single CPU, but using it on a cluster where you can
> distribute the cost for a cooling system across individual units, it
> might appear less crazy.

Possibly. there are a number of people who have done some crazy things
chilling mainboards - check through old slashdot posts - one guy immersed
things in Fluorinert, and it didnt work, and cost him $1000 US odd for
the chemical, another tried mineral oil and a standard A/C compressor,
another alcohol, and I think there are at least a couple LN experiments
out there.

For the money it may well be cheaper to just buy more nodes, wouldn't you
think? Run them at a slightly slower than dangerous speed and you've got more
computing power - you can place these extra nodes in the floor space otherwise
occupied by the chiller for your complex LN system. :)

/kc





More information about the Beowulf mailing list